This week's Statistics Seminar Speaker will be Felix Thoemmes from Cornell University.
Title: Auxiliary Variables in Missing Data Problems: A Graphical Approach
Abstract: The treatment of missing data in the social sciences has changed tremendously during the last decade. Since the seminal work of Donald Rubin, modern missing data techniques such as multiple imputation and full-information maximum likelihood are used much more frequently. These methods assume that data are missing at random (MAR). One very common approach to increase the likelihood that missing at random holds, consists of including many covariates as so-called auxiliary variables. These variables are either included based on data considerations or in an inclusive fashion, i.e., taking all available auxiliary variables. However, neither approach accounts for the fact that there is a class of variables that, when used as auxiliary variables, will always increase bias in the estimation of parameters from data with missing values. In this talk, I will show how graphical models can aid in the identification of such variables. I will present some prototypical situations in which bias due to inclusion of auxiliary variables emerges, and quantify it in simulation studies. Practical approaches on choosing auxiliary variables in applied settings will be discussed.
Refreshments will be served after the seminar in 1181 Comstock Hall.